Hello,
I have several nodes with specific hardware, namely Infiniband. I want all MPI jobs to only be sent to these nodes. To this end, I defined a custom boolean resource “ib”, flag “h”, and set it to true
on all these nodes (and false
on the others). Also added it to the “resources:” line in sched_config
. Indeed, as long as I submit a job with ib=True
it lands on one of the specific nodes.
My next step was to make a dedicated “mpi” queue that will place all submitted jobs on these nodes (without the need to specify ib=True
):
set queue mpi resources_default.ib = True
But this doesn’t work. Instead, I’m getting
09/05/2021 18:18:47 L Insufficient amount of queue resource: ib (True != False) 09/05/2021 18:18:47 L Job will never run with the resources currently configured in the complex
Why??